Abstract. Including a terrestrial nitrogen (N) cycle in Earth system models has led to substantial attenuation of predicted biosphere-climate feedbacks. However, the magnitude of this attenuation remains uncertain. A particularly important but highly uncertain process is biological nitrogen fixation (BNF), which is the largest natural input of N to land ecosystems globally. In order to quantify this uncertainty and estimate likely effects on terrestrial biosphere dynamics, we applied six alternative formulations of BNF spanning the range of process formulations in current stateof-the-art biosphere models within a common framework, the O-CN model: a global map of static BNF rates, two empirical relationships between BNF and other ecosystem variables (net primary productivity and evapotranspiration), two process-oriented formulations based on plant N status, and an optimality-based approach. We examined the resulting differences in model predictions under ambient and elevated atmospheric [CO 2 ] and found that the predicted global BNF rates and their spatial distribution for contemporary conditions were broadly comparable, ranging from 108 to 148 Tg N yr −1 (median: 128 Tg N yr −1 ), despite distinct regional patterns associated with the assumptions of each approach. Notwithstanding, model responses in BNF rates to elevated levels of atmospheric [CO 2 ] (+200 ppm) ranged between −4 Tg N yr −1 (−3 %) and 56 Tg N yr −1 (+42 %) (median: 7 Tg N yr −1 (+8 %)). As a consequence, future projections of global ecosystem carbon (C) storage (+281 to +353 Pg C, or +13 to +16 %) as well as N 2 O emission (−1.6 to +0.5 Tg N yr −1 , or −19 to +7 %) differed significantly across the different model formulations. Our results emphasize the importance of better understanding the nature and magnitude of BNF responses to change-induced perturbations, particularly through new empirical perturbation experiments and improved model representation.

Summary The response of the forest carbon (C) balance to changes in nitrogen (N) deposition is uncertain, partly owing to diverging representations of N cycle processes in dynamic global vegetation models (DGVMs). Here, we examined how different assumptions about the degree of flexibility of the ecosystem's C : N ratios contribute to this uncertainty, and which of these assumptions best correspond to the available data. We applied these assumptions within the framework of a DGVM and compared the results to responses in net primary productivity (NPP), leaf N concentration, and ecosystem N partitioning, observed at 22 forest N fertilization experiments. Employing flexible ecosystem pool C : N ratios generally resulted in the most convincing model–data agreement with respect to production and foliar N responses. An intermediate degree of stoichiometric flexibility in vegetation, where wood C : N ratio changes were decoupled from leaf and root C : N ratio changes, led to consistent simulation of production and N cycle responses to N addition. Assuming fixed C : N ratios or scaling leaf N concentration changes to other tissues, commonly assumed by DGVMs, was not supported by reported data. Between the tested assumptions, the simulated changes in ecosystem C storage relative to changes in C assimilation varied by up to 20%.

Abstract. The nitrogen cycle and its effect on carbon uptake in the terrestrial biosphere is a recent progression in earth system models. As with any new component of a model, it is important to understand the behaviour, strengths, and limitations of the various process representations. Here we assess and compare five models with nitrogen cycles that will be used as the terrestrial components of some of the earth system models in CMIP6. We use a historical control simulation and two perturbations to assess the models' nitrogen-related performance: a simulation with atmospheric carbon dioxide 200 ppm higher, and one with nitrogen deposition increased by 50 kg N ha−1 yr−1. We find that, despite differing nitrogen cycle representations, all models simulate recent global trends in terrestrial productivity and net carbon uptake commensurate with observations. The between-model variation is likely more influenced by other, non-nitrogen parts of the models. Globally, the productivity response to increased carbon dioxide is commensurate with observations for four of the five models, but highly spatially variable within and between models. The productivity response to increased nitrogen is significantly lower than observed in two of the five models. The global and tropical values are generally better represented than boreal, tundra, or other high latitude areas. These results are due to divergent though valid choices in the representation of key processes. They show the need for better understanding and more provision of observational constraints of nitrogen processes, especially nitrogen-use efficiency and biological nitrogen fixation.

Abstract. The nitrogen cycle and its effect on carbon uptake in the terrestrial biosphere is a recent progression in earth system models. As with any new component of a model, it is important to understand the behaviour, strengths, and limitations of the various process representations. Here we assess and compare five land surface models with nitrogen cycles that are used as the terrestrial components of some of the earth system models in CMIP6. The land surface models were run offline with a common spin-up and forcing protocol. We use a historical control simulation and two perturbations to assess the model nitrogen-related performances: a simulation with atmospheric carbon dioxide increased by 200 ppm and one with nitrogen deposition increased by 50 kgN ha−1 yr−1. There is generally greater variability in productivity response between models to increased nitrogen than to carbon dioxide. Across the five models the response to carbon dioxide globally was 5 % to 20 % and the response to nitrogen was 2 % to 24 %. The models are not evenly distributed within the ensemble range, with two of the models having low productivity response to nitrogen and another one with low response to elevated atmospheric carbon dioxide, compared to the other models. In all five models individual grid cells tend to exhibit bimodality, with either a strong response to increased nitrogen or atmospheric carbon dioxide but rarely to both to an equal extent. However, this local effect does not scale to either the regional or global level. The global and tropical responses are generally more accurately modelled than boreal, tundra, or other high-latitude areas compared to observations. These results are due to divergent choices in the representation of key nitrogen cycle processes. They show the need for more observational studies to enhance understanding of nitrogen cycle processes, especially nitrogen-use efficiency and biological nitrogen fixation.

The magnitude of the nitrogen (N) limitation of terrestrial carbon (C) storage over the 21st century is highly uncertain because of the complex interactions between the terrestrial C and N cycles. We use an ensemble approach to quantify and attribute process‐level uncertainty in C‐cycle projections by analysing a 30‐member ensemble representing published alternative representations of key N cycle processes (stoichiometry, biological nitrogen fixation (BNF) and ecosystem N losses) within the framework of one terrestrial biosphere model. Despite large differences in the simulated present‐day N cycle, primarily affecting simulated productivity north of 40°N, ensemble members generally conform with global C‐cycle benchmarks for present‐day conditions. Ensemble projections for two representative concentration pathways (RCP 2.6 and RCP 8.5) show that the increase in land C storage due to CO2 fertilization is reduced by 24 ± 15% due to N constraints, whereas terrestrial C losses associated with climate change are attenuated by 19 ± 20%. As a result, N cycling reduces projected land C uptake for the years 2006–2099 by 19% (37% decrease to 3% increase) for RCP 2.6, and by 21% (40% decrease to 9% increase) for RCP 8.5. Most of the ensemble spread results from uncertainty in temperate and boreal forests, and is dominated by uncertainty in BNF (10% decrease to 50% increase for RCP 2.6, 5% decrease to 100% increase for RCP 8.5). However, choices about the flexibility of ecosystem C:N ratios and processes controlling ecosystem N losses regionally also play important roles. The findings of this study demonstrate clearly the need for an ensemble approach to quantify likely future terrestrial C–N cycle trajectories. Present‐day C‐cycle observations only weakly constrain the future ensemble spread, highlighting the need for better observational constraints on large‐scale N cycling, and N cycle process responses to global change.